def int_x86_avx_maskload_ps_256 : GCCBuiltin<"__builtin_ia32_maskloadps256">,
Intrinsic<[llvm_v8f32_ty], [llvm_ptr_ty, llvm_v8i32_ty],
[IntrReadArgMem]>;
- def int_x86_avx512_mask_loadu_ps_512 : GCCBuiltin<"__builtin_ia32_loadups512_mask">,
- Intrinsic<[llvm_v16f32_ty], [llvm_ptr_ty, llvm_v16f32_ty, llvm_i16_ty],
- [IntrReadArgMem]>;
- def int_x86_avx512_mask_loadu_pd_512 : GCCBuiltin<"__builtin_ia32_loadupd512_mask">,
- Intrinsic<[llvm_v8f64_ty], [llvm_ptr_ty, llvm_v8f64_ty, llvm_i8_ty],
- [IntrReadArgMem]>;
- def int_x86_avx512_mask_load_ps_512 : GCCBuiltin<"__builtin_ia32_loadaps512_mask">,
- Intrinsic<[llvm_v16f32_ty], [llvm_ptr_ty, llvm_v16f32_ty, llvm_i16_ty],
- [IntrReadArgMem]>;
- def int_x86_avx512_mask_load_pd_512 : GCCBuiltin<"__builtin_ia32_loadapd512_mask">,
- Intrinsic<[llvm_v8f64_ty], [llvm_ptr_ty, llvm_v8f64_ty, llvm_i8_ty],
- [IntrReadArgMem]>;
- def int_x86_avx512_mask_move_ss : GCCBuiltin<"__builtin_ia32_movss_mask">,
- Intrinsic<[llvm_v4f32_ty], [llvm_v4f32_ty, llvm_v4f32_ty, llvm_v4f32_ty, llvm_i8_ty],
- [IntrNoMem]>;
- def int_x86_avx512_mask_move_sd : GCCBuiltin<"__builtin_ia32_movsd_mask">,
- Intrinsic<[llvm_v2f64_ty], [llvm_v2f64_ty, llvm_v2f64_ty, llvm_v2f64_ty, llvm_i8_ty],
- [IntrNoMem]>;
+ def int_x86_avx512_mask_loadu_ps_128 :
+ GCCBuiltin<"__builtin_ia32_loadups128_mask">,
+ Intrinsic<[llvm_v4f32_ty],
+ [llvm_ptr_ty, llvm_v4f32_ty, llvm_i8_ty], [IntrReadArgMem]>;
+ def int_x86_avx512_mask_loadu_ps_256 :
+ GCCBuiltin<"__builtin_ia32_loadups256_mask">,
+ Intrinsic<[llvm_v8f32_ty],
+ [llvm_ptr_ty, llvm_v8f32_ty, llvm_i8_ty], [IntrReadArgMem]>;
+ def int_x86_avx512_mask_loadu_ps_512 :
+ GCCBuiltin<"__builtin_ia32_loadups512_mask">,
+ Intrinsic<[llvm_v16f32_ty],
+ [llvm_ptr_ty, llvm_v16f32_ty, llvm_i16_ty], [IntrReadArgMem]>;
+
+ def int_x86_avx512_mask_loadu_pd_128 :
+ GCCBuiltin<"__builtin_ia32_loadupd128_mask">,
+ Intrinsic<[llvm_v2f64_ty],
+ [llvm_ptr_ty, llvm_v2f64_ty, llvm_i8_ty], [IntrReadArgMem]>;
+ def int_x86_avx512_mask_loadu_pd_256 :
+ GCCBuiltin<"__builtin_ia32_loadupd256_mask">,
+ Intrinsic<[llvm_v4f64_ty],
+ [llvm_ptr_ty, llvm_v4f64_ty, llvm_i8_ty], [IntrReadArgMem]>;
+ def int_x86_avx512_mask_loadu_pd_512 :
+ GCCBuiltin<"__builtin_ia32_loadupd512_mask">,
+ Intrinsic<[llvm_v8f64_ty],
+ [llvm_ptr_ty, llvm_v8f64_ty, llvm_i8_ty], [IntrReadArgMem]>;
+
+ def int_x86_avx512_mask_load_ps_128 :
+ GCCBuiltin<"__builtin_ia32_loadaps128_mask">,
+ Intrinsic<[llvm_v4f32_ty],
+ [llvm_ptr_ty, llvm_v4f32_ty, llvm_i8_ty], [IntrReadArgMem]>;
+ def int_x86_avx512_mask_load_ps_256 :
+ GCCBuiltin<"__builtin_ia32_loadaps256_mask">,
+ Intrinsic<[llvm_v8f32_ty],
+ [llvm_ptr_ty, llvm_v8f32_ty, llvm_i8_ty], [IntrReadArgMem]>;
+ def int_x86_avx512_mask_load_ps_512 :
+ GCCBuiltin<"__builtin_ia32_loadaps512_mask">,
+ Intrinsic<[llvm_v16f32_ty],
+ [llvm_ptr_ty, llvm_v16f32_ty, llvm_i16_ty], [IntrReadArgMem]>;
+
+ def int_x86_avx512_mask_load_pd_128 :
+ GCCBuiltin<"__builtin_ia32_loadapd128_mask">,
+ Intrinsic<[llvm_v2f64_ty],
+ [llvm_ptr_ty, llvm_v2f64_ty, llvm_i8_ty], [IntrReadArgMem]>;
+ def int_x86_avx512_mask_load_pd_256 :
+ GCCBuiltin<"__builtin_ia32_loadapd256_mask">,
+ Intrinsic<[llvm_v4f64_ty],
+ [llvm_ptr_ty, llvm_v4f64_ty, llvm_i8_ty], [IntrReadArgMem]>;
+ def int_x86_avx512_mask_load_pd_512 :
+ GCCBuiltin<"__builtin_ia32_loadapd512_mask">,
+ Intrinsic<[llvm_v8f64_ty],
+ [llvm_ptr_ty, llvm_v8f64_ty, llvm_i8_ty], [IntrReadArgMem]>;
+
+ def int_x86_avx512_mask_move_ss :
+ GCCBuiltin<"__builtin_ia32_movss_mask">,
+ Intrinsic<[llvm_v4f32_ty],
+ [llvm_v4f32_ty, llvm_v4f32_ty, llvm_v4f32_ty, llvm_i8_ty],
+ [IntrNoMem]>;
+ def int_x86_avx512_mask_move_sd :
+ GCCBuiltin<"__builtin_ia32_movsd_mask">,
+ Intrinsic<[llvm_v2f64_ty],
+ [llvm_v2f64_ty, llvm_v2f64_ty, llvm_v2f64_ty, llvm_i8_ty],
+ [IntrNoMem]>;
}
// Conditional store ops
}
}
+
+bool X86TargetLowering::getTgtMemIntrinsic(IntrinsicInfo &Info,
+ const CallInst &I,
+ unsigned Intrinsic) const {
+
+ const IntrinsicData* IntrData = getIntrinsicWithChain(Intrinsic);
+ if (!IntrData)
+ return false;
+
+ switch (IntrData->Type) {
+ case LOADA:
+ case LOADU: {
+ Info.opc = ISD::INTRINSIC_W_CHAIN;
+ Info.memVT = MVT::getVT(I.getType());
+ Info.ptrVal = I.getArgOperand(0);
+ Info.offset = 0;
+ Info.align = (IntrData->Type == LOADA ? Info.memVT.getSizeInBits()/8 : 1);
+ Info.vol = false;
+ Info.readMem = true;
+ Info.writeMem = false;
+ return true;
+ }
+ default:
+ break;
+ }
+
+ return false;
+}
+
/// Returns true if the target can instruction select the
/// specified FP immediate natively. If false, the legalizer will
/// materialize the FP immediate as a load from a constant pool.
return DAG.getMergeValues(Results, dl);
}
case COMPRESS_TO_MEM: {
- SDLoc dl(Op);
SDValue Mask = Op.getOperand(4);
SDValue DataToCompress = Op.getOperand(3);
SDValue Addr = Op.getOperand(2);
case TRUNCATE_TO_MEM_VI32:
return LowerINTRINSIC_TRUNCATE_TO_MEM(Op, DAG, MVT::i32);
case EXPAND_FROM_MEM: {
- SDLoc dl(Op);
SDValue Mask = Op.getOperand(4);
SDValue PassThru = Op.getOperand(3);
SDValue Addr = Op.getOperand(2);
Mask, PassThru, Subtarget, DAG), Chain};
return DAG.getMergeValues(Results, dl);
}
+ case LOADU:
+ case LOADA: {
+ SDValue Mask = Op.getOperand(4);
+ SDValue PassThru = Op.getOperand(3);
+ SDValue Addr = Op.getOperand(2);
+ SDValue Chain = Op.getOperand(0);
+ MVT VT = Op.getSimpleValueType();
+
+ MemIntrinsicSDNode *MemIntr = dyn_cast<MemIntrinsicSDNode>(Op);
+ assert(MemIntr && "Expected MemIntrinsicSDNode!");
+
+ if (isAllOnesConstant(Mask)) // return just a load
+ return DAG.getLoad(VT, dl, Chain, Addr, MemIntr->getMemOperand());
+
+ MVT MaskVT = MVT::getVectorVT(MVT::i1, VT.getVectorNumElements());
+ SDValue VMask = getMaskNode(Mask, MaskVT, Subtarget, DAG, dl);
+ return DAG.getMaskedLoad(VT, dl, Chain, Addr, VMask, PassThru, VT,
+ MemIntr->getMemOperand(), ISD::NON_EXTLOAD);
+ }
}
}
/// from i32 to i8 but not from i32 to i16.
bool isNarrowingProfitable(EVT VT1, EVT VT2) const override;
+ /// Given an intrinsic, checks if on the target the intrinsic will need to map
+ /// to a MemIntrinsicNode (touches memory). If this is the case, it returns
+ /// true and stores the intrinsic information into the IntrinsicInfo that was
+ /// passed to the function.
+ bool getTgtMemIntrinsic(IntrinsicInfo &Info, const CallInst &I,
+ unsigned Intrinsic) const override;
+
/// Returns true if the target can instruction select the
/// specified FP immediate natively. If false, the legalizer will
/// materialize the FP immediate as a load from a constant pool.
avx512_store_vl<0x11, "vmovupd", avx512vl_f64_info, HasAVX512>,
PD, VEX_W, EVEX_CD8<64, CD8VF>;
-def: Pat<(v8f64 (int_x86_avx512_mask_loadu_pd_512 addr:$ptr,
- (bc_v8f64 (v16i32 immAllZerosV)), GR8:$mask)),
- (VMOVUPDZrmkz (v8i1 (COPY_TO_REGCLASS GR8:$mask, VK8WM)), addr:$ptr)>;
-
-def: Pat<(v16f32 (int_x86_avx512_mask_loadu_ps_512 addr:$ptr,
- (bc_v16f32 (v16i32 immAllZerosV)), GR16:$mask)),
- (VMOVUPSZrmkz (v16i1 (COPY_TO_REGCLASS GR16:$mask, VK16WM)), addr:$ptr)>;
-
-def: Pat<(v8f64 (int_x86_avx512_mask_load_pd_512 addr:$ptr,
- (bc_v8f64 (v16i32 immAllZerosV)), GR8:$mask)),
- (VMOVAPDZrmkz (v8i1 (COPY_TO_REGCLASS GR8:$mask, VK8WM)), addr:$ptr)>;
-
-def: Pat<(v16f32 (int_x86_avx512_mask_load_ps_512 addr:$ptr,
- (bc_v16f32 (v16i32 immAllZerosV)), GR16:$mask)),
- (VMOVAPSZrmkz (v16i1 (COPY_TO_REGCLASS GR16:$mask, VK16WM)), addr:$ptr)>;
-
-def: Pat<(v8f64 (int_x86_avx512_mask_load_pd_512 addr:$ptr,
- (bc_v8f64 (v16i32 immAllZerosV)), (i8 -1))),
- (VMOVAPDZrm addr:$ptr)>;
-
-def: Pat<(v16f32 (int_x86_avx512_mask_load_ps_512 addr:$ptr,
- (bc_v16f32 (v16i32 immAllZerosV)), (i16 -1))),
- (VMOVAPSZrm addr:$ptr)>;
-
def: Pat<(int_x86_avx512_mask_storeu_ps_512 addr:$ptr, (v16f32 VR512:$src),
GR16:$mask),
(VMOVUPSZmrk addr:$ptr, (v16i1 (COPY_TO_REGCLASS GR16:$mask, VK16WM)),
INTR_TYPE_SCALAR_MASK_RM, INTR_TYPE_3OP_SCALAR_MASK_RM,
COMPRESS_EXPAND_IN_REG, COMPRESS_TO_MEM, BRCST_SUBVEC_TO_VEC,
TRUNCATE_TO_MEM_VI8, TRUNCATE_TO_MEM_VI16, TRUNCATE_TO_MEM_VI32,
- EXPAND_FROM_MEM, BLEND, INSERT_SUBVEC,
+ EXPAND_FROM_MEM, LOADA, LOADU, BLEND, INSERT_SUBVEC,
TERLOG_OP_MASK, TERLOG_OP_MASKZ, BROADCASTM, KUNPCK, CONVERT_MASK_TO_VEC, CONVERT_TO_MASK
};
EXPAND_FROM_MEM, X86ISD::EXPAND, 0),
X86_INTRINSIC_DATA(avx512_mask_expand_load_q_512,
EXPAND_FROM_MEM, X86ISD::EXPAND, 0),
+ X86_INTRINSIC_DATA(avx512_mask_load_pd_128, LOADA, ISD::DELETED_NODE, 0),
+ X86_INTRINSIC_DATA(avx512_mask_load_pd_256, LOADA, ISD::DELETED_NODE, 0),
+ X86_INTRINSIC_DATA(avx512_mask_load_pd_512, LOADA, ISD::DELETED_NODE, 0),
+ X86_INTRINSIC_DATA(avx512_mask_load_ps_128, LOADA, ISD::DELETED_NODE, 0),
+ X86_INTRINSIC_DATA(avx512_mask_load_ps_256, LOADA, ISD::DELETED_NODE, 0),
+ X86_INTRINSIC_DATA(avx512_mask_load_ps_512, LOADA, ISD::DELETED_NODE, 0),
+ X86_INTRINSIC_DATA(avx512_mask_loadu_pd_128, LOADU, ISD::DELETED_NODE, 0),
+ X86_INTRINSIC_DATA(avx512_mask_loadu_pd_256, LOADU, ISD::DELETED_NODE, 0),
+ X86_INTRINSIC_DATA(avx512_mask_loadu_pd_512, LOADU, ISD::DELETED_NODE, 0),
+ X86_INTRINSIC_DATA(avx512_mask_loadu_ps_128, LOADU, ISD::DELETED_NODE, 0),
+ X86_INTRINSIC_DATA(avx512_mask_loadu_ps_256, LOADU, ISD::DELETED_NODE, 0),
+ X86_INTRINSIC_DATA(avx512_mask_loadu_ps_512, LOADU, ISD::DELETED_NODE, 0),
X86_INTRINSIC_DATA(avx512_mask_pmov_db_mem_128, TRUNCATE_TO_MEM_VI8,
X86ISD::VTRUNC, 0),
X86_INTRINSIC_DATA(avx512_mask_pmov_db_mem_256, TRUNCATE_TO_MEM_VI8,
declare void @llvm.x86.avx512.mask.store.pd.512(i8*, <8 x double>, i8)
-define <16 x float> @test_maskz_load_aligned_ps(<16 x float> %data, i8* %ptr, i16 %mask) {
-; CHECK-LABEL: test_maskz_load_aligned_ps:
+define <16 x float> @test_mask_load_aligned_ps(<16 x float> %data, i8* %ptr, i16 %mask) {
+; CHECK-LABEL: test_mask_load_aligned_ps:
; CHECK: ## BB#0:
; CHECK-NEXT: kmovw %esi, %k1
-; CHECK-NEXT: vmovaps (%rdi), %zmm0 {%k1} {z}
+; CHECK-NEXT: vmovaps (%rdi), %zmm0
+; CHECK-NEXT: vmovaps (%rdi), %zmm0 {%k1}
+; CHECK-NEXT: vmovaps (%rdi), %zmm1 {%k1} {z}
+; CHECK-NEXT: vaddps %zmm0, %zmm1, %zmm0
; CHECK-NEXT: retq
- %res = call <16 x float> @llvm.x86.avx512.mask.load.ps.512(i8* %ptr, <16 x float> zeroinitializer, i16 %mask)
- ret <16 x float> %res
+ %res = call <16 x float> @llvm.x86.avx512.mask.load.ps.512(i8* %ptr, <16 x float> zeroinitializer, i16 -1)
+ %res1 = call <16 x float> @llvm.x86.avx512.mask.load.ps.512(i8* %ptr, <16 x float> %res, i16 %mask)
+ %res2 = call <16 x float> @llvm.x86.avx512.mask.load.ps.512(i8* %ptr, <16 x float> zeroinitializer, i16 %mask)
+ %res4 = fadd <16 x float> %res2, %res1
+ ret <16 x float> %res4
}
declare <16 x float> @llvm.x86.avx512.mask.load.ps.512(i8*, <16 x float>, i16)
-define <8 x double> @test_maskz_load_aligned_pd(<8 x double> %data, i8* %ptr, i8 %mask) {
-; CHECK-LABEL: test_maskz_load_aligned_pd:
+define <16 x float> @test_mask_load_unaligned_ps(<16 x float> %data, i8* %ptr, i16 %mask) {
+; CHECK-LABEL: test_mask_load_unaligned_ps:
; CHECK: ## BB#0:
; CHECK-NEXT: kmovw %esi, %k1
-; CHECK-NEXT: vmovapd (%rdi), %zmm0 {%k1} {z}
+; CHECK-NEXT: vmovups (%rdi), %zmm0
+; CHECK-NEXT: vmovups (%rdi), %zmm0 {%k1}
+; CHECK-NEXT: vmovups (%rdi), %zmm1 {%k1} {z}
+; CHECK-NEXT: vaddps %zmm0, %zmm1, %zmm0
; CHECK-NEXT: retq
- %res = call <8 x double> @llvm.x86.avx512.mask.load.pd.512(i8* %ptr, <8 x double> zeroinitializer, i8 %mask)
- ret <8 x double> %res
+ %res = call <16 x float> @llvm.x86.avx512.mask.loadu.ps.512(i8* %ptr, <16 x float> zeroinitializer, i16 -1)
+ %res1 = call <16 x float> @llvm.x86.avx512.mask.loadu.ps.512(i8* %ptr, <16 x float> %res, i16 %mask)
+ %res2 = call <16 x float> @llvm.x86.avx512.mask.loadu.ps.512(i8* %ptr, <16 x float> zeroinitializer, i16 %mask)
+ %res4 = fadd <16 x float> %res2, %res1
+ ret <16 x float> %res4
}
-declare <8 x double> @llvm.x86.avx512.mask.load.pd.512(i8*, <8 x double>, i8)
+declare <16 x float> @llvm.x86.avx512.mask.loadu.ps.512(i8*, <16 x float>, i16)
-define <16 x float> @test_load_aligned_ps(<16 x float> %data, i8* %ptr, i16 %mask) {
-; CHECK-LABEL: test_load_aligned_ps:
+define <8 x double> @test_mask_load_aligned_pd(<8 x double> %data, i8* %ptr, i8 %mask) {
+; CHECK-LABEL: test_mask_load_aligned_pd:
; CHECK: ## BB#0:
-; CHECK-NEXT: vmovaps (%rdi), %zmm0
+; CHECK-NEXT: movzbl %sil, %eax
+; CHECK-NEXT: kmovw %eax, %k1
+; CHECK-NEXT: vmovapd (%rdi), %zmm0
+; CHECK-NEXT: vmovapd (%rdi), %zmm0 {%k1}
+; CHECK-NEXT: vmovapd (%rdi), %zmm1 {%k1} {z}
+; CHECK-NEXT: vaddpd %zmm0, %zmm1, %zmm0
; CHECK-NEXT: retq
- %res = call <16 x float> @llvm.x86.avx512.mask.load.ps.512(i8* %ptr, <16 x float> zeroinitializer, i16 -1)
- ret <16 x float> %res
+ %res = call <8 x double> @llvm.x86.avx512.mask.load.pd.512(i8* %ptr, <8 x double> zeroinitializer, i8 -1)
+ %res1 = call <8 x double> @llvm.x86.avx512.mask.load.pd.512(i8* %ptr, <8 x double> %res, i8 %mask)
+ %res2 = call <8 x double> @llvm.x86.avx512.mask.load.pd.512(i8* %ptr, <8 x double> zeroinitializer, i8 %mask)
+ %res4 = fadd <8 x double> %res2, %res1
+ ret <8 x double> %res4
}
-define <8 x double> @test_load_aligned_pd(<8 x double> %data, i8* %ptr, i8 %mask) {
-; CHECK-LABEL: test_load_aligned_pd:
+declare <8 x double> @llvm.x86.avx512.mask.load.pd.512(i8*, <8 x double>, i8)
+
+define <8 x double> @test_mask_load_unaligned_pd(<8 x double> %data, i8* %ptr, i8 %mask) {
+; CHECK-LABEL: test_mask_load_unaligned_pd:
; CHECK: ## BB#0:
-; CHECK-NEXT: vmovapd (%rdi), %zmm0
+; CHECK-NEXT: movzbl %sil, %eax
+; CHECK-NEXT: kmovw %eax, %k1
+; CHECK-NEXT: vmovupd (%rdi), %zmm0
+; CHECK-NEXT: vmovupd (%rdi), %zmm0 {%k1}
+; CHECK-NEXT: vmovupd (%rdi), %zmm1 {%k1} {z}
+; CHECK-NEXT: vaddpd %zmm0, %zmm1, %zmm0
; CHECK-NEXT: retq
- %res = call <8 x double> @llvm.x86.avx512.mask.load.pd.512(i8* %ptr, <8 x double> zeroinitializer, i8 -1)
- ret <8 x double> %res
+ %res = call <8 x double> @llvm.x86.avx512.mask.loadu.pd.512(i8* %ptr, <8 x double> zeroinitializer, i8 -1)
+ %res1 = call <8 x double> @llvm.x86.avx512.mask.loadu.pd.512(i8* %ptr, <8 x double> %res, i8 %mask)
+ %res2 = call <8 x double> @llvm.x86.avx512.mask.loadu.pd.512(i8* %ptr, <8 x double> zeroinitializer, i8 %mask)
+ %res4 = fadd <8 x double> %res2, %res1
+ ret <8 x double> %res4
}
-declare <8 x i64> @llvm.x86.avx512.movntdqa(i8*)
+declare <8 x double> @llvm.x86.avx512.mask.loadu.pd.512(i8*, <8 x double>, i8)
define <8 x i64> @test_valign_q(<8 x i64> %a, <8 x i64> %b) {
; CHECK-LABEL: test_valign_q:
ret <4 x i64> %res4
}
+define <8 x float> @test_mask_load_aligned_ps_256(<8 x float> %data, i8* %ptr, i8 %mask) {
+; CHECK-LABEL: test_mask_load_aligned_ps_256:
+; CHECK: ## BB#0:
+; CHECK-NEXT: movzbl %sil, %eax
+; CHECK-NEXT: kmovw %eax, %k1
+; CHECK-NEXT: vmovaps (%rdi), %ymm0
+; CHECK-NEXT: vmovaps (%rdi), %ymm0 {%k1}
+; CHECK-NEXT: vmovaps (%rdi), %ymm1 {%k1} {z}
+; CHECK-NEXT: vaddps %ymm0, %ymm1, %ymm0
+; CHECK-NEXT: retq
+ %res = call <8 x float> @llvm.x86.avx512.mask.load.ps.256(i8* %ptr, <8 x float> zeroinitializer, i8 -1)
+ %res1 = call <8 x float> @llvm.x86.avx512.mask.load.ps.256(i8* %ptr, <8 x float> %res, i8 %mask)
+ %res2 = call <8 x float> @llvm.x86.avx512.mask.load.ps.256(i8* %ptr, <8 x float> zeroinitializer, i8 %mask)
+ %res4 = fadd <8 x float> %res2, %res1
+ ret <8 x float> %res4
+}
+
+declare <8 x float> @llvm.x86.avx512.mask.load.ps.256(i8*, <8 x float>, i8)
+
+define <8 x float> @test_mask_load_unaligned_ps_256(<8 x float> %data, i8* %ptr, i8 %mask) {
+; CHECK-LABEL: test_mask_load_unaligned_ps_256:
+; CHECK: ## BB#0:
+; CHECK-NEXT: movzbl %sil, %eax
+; CHECK-NEXT: kmovw %eax, %k1
+; CHECK-NEXT: vmovups (%rdi), %ymm0
+; CHECK-NEXT: vmovups (%rdi), %ymm0 {%k1}
+; CHECK-NEXT: vmovups (%rdi), %ymm1 {%k1} {z}
+; CHECK-NEXT: vaddps %ymm0, %ymm1, %ymm0
+; CHECK-NEXT: retq
+ %res = call <8 x float> @llvm.x86.avx512.mask.loadu.ps.256(i8* %ptr, <8 x float> zeroinitializer, i8 -1)
+ %res1 = call <8 x float> @llvm.x86.avx512.mask.loadu.ps.256(i8* %ptr, <8 x float> %res, i8 %mask)
+ %res2 = call <8 x float> @llvm.x86.avx512.mask.loadu.ps.256(i8* %ptr, <8 x float> zeroinitializer, i8 %mask)
+ %res4 = fadd <8 x float> %res2, %res1
+ ret <8 x float> %res4
+}
+
+declare <8 x float> @llvm.x86.avx512.mask.loadu.ps.256(i8*, <8 x float>, i8)
+
+define <4 x double> @test_mask_load_aligned_pd_256(<4 x double> %data, i8* %ptr, i8 %mask) {
+; CHECK-LABEL: test_mask_load_aligned_pd_256:
+; CHECK: ## BB#0:
+; CHECK-NEXT: movzbl %sil, %eax
+; CHECK-NEXT: kmovw %eax, %k1
+; CHECK-NEXT: vmovapd (%rdi), %ymm0
+; CHECK-NEXT: vmovapd (%rdi), %ymm0 {%k1}
+; CHECK-NEXT: vmovapd (%rdi), %ymm1 {%k1} {z}
+; CHECK-NEXT: vaddpd %ymm0, %ymm1, %ymm0
+; CHECK-NEXT: retq
+ %res = call <4 x double> @llvm.x86.avx512.mask.load.pd.256(i8* %ptr, <4 x double> zeroinitializer, i8 -1)
+ %res1 = call <4 x double> @llvm.x86.avx512.mask.load.pd.256(i8* %ptr, <4 x double> %res, i8 %mask)
+ %res2 = call <4 x double> @llvm.x86.avx512.mask.load.pd.256(i8* %ptr, <4 x double> zeroinitializer, i8 %mask)
+ %res4 = fadd <4 x double> %res2, %res1
+ ret <4 x double> %res4
+}
+
+declare <4 x double> @llvm.x86.avx512.mask.load.pd.256(i8*, <4 x double>, i8)
+
+define <4 x double> @test_mask_load_unaligned_pd_256(<4 x double> %data, i8* %ptr, i8 %mask) {
+; CHECK-LABEL: test_mask_load_unaligned_pd_256:
+; CHECK: ## BB#0:
+; CHECK-NEXT: movzbl %sil, %eax
+; CHECK-NEXT: kmovw %eax, %k1
+; CHECK-NEXT: vmovupd (%rdi), %ymm0
+; CHECK-NEXT: vmovupd (%rdi), %ymm0 {%k1}
+; CHECK-NEXT: vmovupd (%rdi), %ymm1 {%k1} {z}
+; CHECK-NEXT: vaddpd %ymm0, %ymm1, %ymm0
+; CHECK-NEXT: retq
+ %res = call <4 x double> @llvm.x86.avx512.mask.loadu.pd.256(i8* %ptr, <4 x double> zeroinitializer, i8 -1)
+ %res1 = call <4 x double> @llvm.x86.avx512.mask.loadu.pd.256(i8* %ptr, <4 x double> %res, i8 %mask)
+ %res2 = call <4 x double> @llvm.x86.avx512.mask.loadu.pd.256(i8* %ptr, <4 x double> zeroinitializer, i8 %mask)
+ %res4 = fadd <4 x double> %res2, %res1
+ ret <4 x double> %res4
+}
+
+declare <4 x double> @llvm.x86.avx512.mask.loadu.pd.256(i8*, <4 x double>, i8)
+
+define <4 x float> @test_mask_load_aligned_ps_128(<4 x float> %data, i8* %ptr, i8 %mask) {
+; CHECK-LABEL: test_mask_load_aligned_ps_128:
+; CHECK: ## BB#0:
+; CHECK-NEXT: movzbl %sil, %eax
+; CHECK-NEXT: kmovw %eax, %k1
+; CHECK-NEXT: vmovaps (%rdi), %xmm0
+; CHECK-NEXT: vmovaps (%rdi), %xmm0 {%k1}
+; CHECK-NEXT: vmovaps (%rdi), %xmm1 {%k1} {z}
+; CHECK-NEXT: vaddps %xmm0, %xmm1, %xmm0
+; CHECK-NEXT: retq
+ %res = call <4 x float> @llvm.x86.avx512.mask.load.ps.128(i8* %ptr, <4 x float> zeroinitializer, i8 -1)
+ %res1 = call <4 x float> @llvm.x86.avx512.mask.load.ps.128(i8* %ptr, <4 x float> %res, i8 %mask)
+ %res2 = call <4 x float> @llvm.x86.avx512.mask.load.ps.128(i8* %ptr, <4 x float> zeroinitializer, i8 %mask)
+ %res4 = fadd <4 x float> %res2, %res1
+ ret <4 x float> %res4
+}
+
+declare <4 x float> @llvm.x86.avx512.mask.load.ps.128(i8*, <4 x float>, i8)
+
+define <4 x float> @test_mask_load_unaligned_ps_128(<4 x float> %data, i8* %ptr, i8 %mask) {
+; CHECK-LABEL: test_mask_load_unaligned_ps_128:
+; CHECK: ## BB#0:
+; CHECK-NEXT: movzbl %sil, %eax
+; CHECK-NEXT: kmovw %eax, %k1
+; CHECK-NEXT: vmovups (%rdi), %xmm0
+; CHECK-NEXT: vmovups (%rdi), %xmm0 {%k1}
+; CHECK-NEXT: vmovups (%rdi), %xmm1 {%k1} {z}
+; CHECK-NEXT: vaddps %xmm0, %xmm1, %xmm0
+; CHECK-NEXT: retq
+ %res = call <4 x float> @llvm.x86.avx512.mask.loadu.ps.128(i8* %ptr, <4 x float> zeroinitializer, i8 -1)
+ %res1 = call <4 x float> @llvm.x86.avx512.mask.loadu.ps.128(i8* %ptr, <4 x float> %res, i8 %mask)
+ %res2 = call <4 x float> @llvm.x86.avx512.mask.loadu.ps.128(i8* %ptr, <4 x float> zeroinitializer, i8 %mask)
+ %res4 = fadd <4 x float> %res2, %res1
+ ret <4 x float> %res4
+}
+
+declare <4 x float> @llvm.x86.avx512.mask.loadu.ps.128(i8*, <4 x float>, i8)
+
+define <2 x double> @test_mask_load_aligned_pd_128(<2 x double> %data, i8* %ptr, i8 %mask) {
+; CHECK-LABEL: test_mask_load_aligned_pd_128:
+; CHECK: ## BB#0:
+; CHECK-NEXT: movzbl %sil, %eax
+; CHECK-NEXT: kmovw %eax, %k1
+; CHECK-NEXT: vmovapd (%rdi), %xmm0
+; CHECK-NEXT: vmovapd (%rdi), %xmm0 {%k1}
+; CHECK-NEXT: vmovapd (%rdi), %xmm1 {%k1} {z}
+; CHECK-NEXT: vaddpd %xmm0, %xmm1, %xmm0
+; CHECK-NEXT: retq
+ %res = call <2 x double> @llvm.x86.avx512.mask.load.pd.128(i8* %ptr, <2 x double> zeroinitializer, i8 -1)
+ %res1 = call <2 x double> @llvm.x86.avx512.mask.load.pd.128(i8* %ptr, <2 x double> %res, i8 %mask)
+ %res2 = call <2 x double> @llvm.x86.avx512.mask.load.pd.128(i8* %ptr, <2 x double> zeroinitializer, i8 %mask)
+ %res4 = fadd <2 x double> %res2, %res1
+ ret <2 x double> %res4
+}
+
+declare <2 x double> @llvm.x86.avx512.mask.load.pd.128(i8*, <2 x double>, i8)
+
+define <2 x double> @test_mask_load_unaligned_pd_128(<2 x double> %data, i8* %ptr, i8 %mask) {
+; CHECK-LABEL: test_mask_load_unaligned_pd_128:
+; CHECK: ## BB#0:
+; CHECK-NEXT: movzbl %sil, %eax
+; CHECK-NEXT: kmovw %eax, %k1
+; CHECK-NEXT: vmovupd (%rdi), %xmm0
+; CHECK-NEXT: vmovupd (%rdi), %xmm0 {%k1}
+; CHECK-NEXT: vmovupd (%rdi), %xmm1 {%k1} {z}
+; CHECK-NEXT: vaddpd %xmm0, %xmm1, %xmm0
+; CHECK-NEXT: retq
+ %res = call <2 x double> @llvm.x86.avx512.mask.loadu.pd.128(i8* %ptr, <2 x double> zeroinitializer, i8 -1)
+ %res1 = call <2 x double> @llvm.x86.avx512.mask.loadu.pd.128(i8* %ptr, <2 x double> %res, i8 %mask)
+ %res2 = call <2 x double> @llvm.x86.avx512.mask.loadu.pd.128(i8* %ptr, <2 x double> zeroinitializer, i8 %mask)
+ %res4 = fadd <2 x double> %res2, %res1
+ ret <2 x double> %res4
+}
+
+declare <2 x double> @llvm.x86.avx512.mask.loadu.pd.128(i8*, <2 x double>, i8)
+
declare <4 x i32> @llvm.x86.avx512.mask.psrav4.si(<4 x i32>, <4 x i32>, <4 x i32>, i8)
define <4 x i32>@test_int_x86_avx512_mask_psrav4_si(<4 x i32> %x0, <4 x i32> %x1, <4 x i32> %x2, i8 %x3) {